Optimization Of Non-shared Storage Virtual Machine Live Migration System Based On Ceph

With the development of cloud computing and virtualization,the application of virtual machine live migration has become very widespread.At the same time,due to the improvement of cloud storage,virtual machines begin to use distributed storage as disk data,that is,image file.One of the most popular distributed storage systems is the Ceph block device.In certain circumstances,there is no shared storage.Therefore,during the virtual machine live migration,the disk data must also be transferred.Since the volume of disk data is very large,the impact on migration efficiency is very significant.Therefore,how to optimize the migration time of disk data has become a research hotspot.The main task of this thesis is to study and deploy a non-shared storage virtual machine live migration system based on Ceph in a campus network,and then propose a strategy to optimize the migration time of this system.This thesis introduces related knowledge and technology of virtual machine live migration and Ceph,explains and analyzes the existing virtual machine live migration strategies,and points out the advantages and disadvantages of these strategies.Combined with the technical features of Ceph and the characteristics that each node in the campus network can directly communicate with each other,this thesis proposes a migration algorithm for Ceph block device cross clusters.Image data go from the source storage nodes to the destination storage nodes in parallel,which uses the network of storage nodes.Then this thesis analyzes some factors that affect the efficiency of this algorithm and verified them from experiments.This thesis deploys a non-shared storage virtual machine live migration system based on Ceph in the campus network environment,using QEMU-KVM as a virtual machine manager.This system uses the migration algorithm for Ceph block device cross clusters,all-zero blocks ignored strategy and dirty blocks transferred parallel strategy to optimize the migration.Then this thesis does some experiments to test the performance of the new system.The migration algorithm for Ceph block device cross clusters and the dirty blocks transferred parallel strategy optimize the problem of how to transfer data blocks and speeds up the migration.The all-zero blocks ignored strategy optimizes the problem of which data blocks are transferred and reduces the amount oftransferred data.The result of the experiments shows that,compared with the traditional migration systems,this virtual machine live migration system greatly shortens the migration time and effectively improves the migration efficiency.